Background Currently there is no ARDS definition or classification system that allows optimal prediction of mortality in ARDS patients. This study aimed to examine the predictive values of the AECC and Berlin definitions, as well as clinical and respiratory parameters obtained at onset of ARDS and in the course of the first seven consecutive days. Methods The observational study was conducted at a 14-bed intensive care unit specialized on treatment of ARDS. Predictive validity of the AECC and Berlin definitions as well as PaO2/FiO2 and FiO2/PaO2*Pmean (oxygenation index) on mortality of ARDS patients was assessed and statistically compared. Results Four hundred forty two critically-ill patients admitted for ARDS were analysed. Multivariate Cox regression indicated that the oxygenation index was the most accurate parameter for mortality prediction. The third day after ARDS criteria were met at our hospital was found to represent the best compromise between earliness and accuracy of prognosis of mortality regarding the time of assessment. An oxygenation index of 15 or greater was associated with higher mortality, longer length of stay in ICU and hospital and longer duration of mechanical ventilation. In addition, non-survivors had a significantly longer length of stay and duration of mechanical ventilation in referring hospitals before admitted to the national reference centre than survivors. Conclusions The oxygenation index is suggested to be the most suitable parameter to predict mortality in ARDS, preferably assessed on day 3 after admission to a specialized centre. Patients might benefit when transferred to specialized ICU centres as soon as possible for further treatment.